octoberxin-critique

name: octoberxin-critique version: 1.0.0 description: "5-stage intelligence critique pipeline. Multi-agent critique system for deep analysis, verification, and synthesis of research outputs." author: OctoberXin (Research Specialist) keywords: [critique, analysis, verification, synthesis, multi-agent, research, pipeline, octoberxin] metadata: hermes: emoji: "🔬" depends: - research-pipeline - multi-agent-swarm - cost-router triggers: - "research completion" - "content verification request" - "deep analysis required" - "critique mode" model_tier: "T2/T3"

PLATFORM_MAPPING

OpenClaw → Hermes — All ~/.openclaw/ paths and openclaw CLI commands in this document map to ~/.hermes/ and hermes / delegate_task / todo equivalents. See the root SKILL.md for the full mapping table.

OctoberXin Critique 🔬

5-stage intelligence critique pipeline. A multi-agent system designed for rigorous analysis, verification, and synthesis of research outputs through specialized critique agents.

Overview

1. OctPortal — Receives and routes incoming research content
2. OctMine — Extracts key insights, claims, and evidence
3. OctJudge — Prioritizes and scores critical elements
4. OctSkeptic — Verifies claims against sources and logic
5. OctWeave — Synthesizes final polished critique output

The 5-Stage Pipeline

┌──────────┐     ┌──────────┐     ┌──────────┐     ┌──────────┐     ┌──────────┐
│  Oct     │────→│  Oct     │────→│  Oct     │────→│  Oct     │────→│  Oct     │
│  Portal  │     │  Mine    │     │  Judge   │     │ Skeptic  │     │  Weave   │
└──────────┘     └──────────┘     └──────────┘     └──────────┘     └──────────┘
     │                 │                 │                 │                 │
  Receipt         Extraction      Prioritization    Verification     Synthesis
     │                 │                 │                 │                 │
  Ingest          Pull out         Rank by         Fact-check        Weave
  content          insights        importance       & validate        final
                                                                        output

Stage 1: Receipt (OctPortal)

• Ingests research content, documents, or claims
• Validates input format and completeness
• Routes to appropriate downstream agents
• Creates audit trail for traceability

Stage 2: Extraction (OctMine)

• Identifies key claims and assertions
• Extracts supporting evidence and sources
• Maps dependencies between claims
• Flags missing or weak evidence

Stage 3: Prioritization (OctJudge)

• Scores claims by impact and risk
• Ranks evidence strength
• Identifies critical path items
• Allocates verification resources

Stage 4: Verification (OctSkeptic)

• Cross-references sources
• Identifies logical fallacies
• Tests assumptions
• Flags contradictions

Stage 5: Synthesis (OctWeave)

• Combines verified findings
• Generates structured critique report
• Provides actionable recommendations
• Outputs final intelligence summary

Usage

Trigger Full Pipeline

# Run complete 5-stage critique on research output
critique_result = await run_critique_pipeline(
    content=research_output,
    depth="comprehensive",  # or "quick", "standard"
    focus_areas=["factual_accuracy", "logical_consistency", "source_quality"]
)

Individual Stage Invocation

# Run specific stage only
portal_result = await oct_portal.ingest(content)
mine_result = await oct_mine.extract(portal_result)
judge_result = await oct_judge.prioritize(mine_result)
skeptic_result = await oct_skeptic.verify(judge_result)
weave_result = await oct_weave.synthesize(skeptic_result)

Command Line

# Full pipeline critique
delegate_task with critique skill + adversarial-stress-test --input research.md --output critique.md

# Quick critique (stages 1, 4, 5 only)
delegate_task with critique skill + adversarial-stress-test --input research.md --mode quick

# Deep critique (all stages with recursion)
delegate_task with critique skill + adversarial-stress-test --input research.md --mode deep --recursive

Agent Specifications

Output Format

# Critique Report: [Subject]

## Executive Summary
[High-level findings in 2-3 sentences]

## Critical Findings
1. **[Severity]** Finding description
   - Evidence: [source]
   - Impact: [consequence]
   - Recommendation: [action]

## Verified Claims
- ✅ [Claim with strong evidence]
- ⚠️ [Claim with partial evidence]

## Red Flags
- 🚩 [Critical issue requiring attention]

## Recommendations
1. [Priority 1 recommendation]
2. [Priority 2 recommendation]

## Confidence Score: [X.X]/5.0

Configuration

{
  "octoberxinCritique": {
    "enabled": true,
    "defaultDepth": "standard",
    "recursiveVerification": false,
    "parallelStages": ["OctMine", "OctJudge"],
    "minConfidenceThreshold": 3.0,
    "logToMemory": true,
    "generateDiff": true
  }
}

Best Practices

1. Use full pipeline for high-stakes research validation
2. Use quick mode for rapid sanity checks
3. Enable recursive mode when dealing with complex nested claims
4. Always review OctSkeptic output — it's the quality gate
5. Log all critiques to memory for pattern analysis
6. Compare against historical critiques to identify recurring issues

Integration Points

• OctoberXin: Primary user, research specialist
• Halloween: Code/architecture review mode
• October: Coordination and routing
• Memory System: Stores all critique outputs for retrieval
• Cost Router: Manages model tier allocation per stage

Limitations

• Verification limited by source accessibility
• Cannot verify claims without sufficient evidence
• Recursive mode increases token usage significantly
• Best suited for analytical content; less effective for creative work